An experimentation of objective functions used for multilevel thresholding based image segmentation using particle swarm optimization

For image segmentation, multilevel thresholding is treated as one of the widely used approach. However, this approach has a major issue that is it suffers of high computational complexity problem with the increase threshold levels. This paper presented a comparative performance analysis of some objective functions used for image segmentation using the concept of multilevel thresholding and a modified version of adaptive inertia weight Particle Swarm Optimization (PSO) technique. The PSO algorithm is applied to multilevel thresholding based image segmentation using either Otsu’s inter class variance, Kapur’s entropy or Masi entropy as an objective function. Each method is tested over various standard image dataset like Berkeley image database, USC-SIPI image dataset etc. The evaluated result of each method has been compared and the overall experimentation is performed in three different ways such as, PSO and Otsu’s method, PSO and Kapur’s entropy and PSO with Masi entropy. The performance and quality of the segmented images are measured using the following parameters such as, average Mean Structural Similarity Index (MSSIM), average Peak Signal to Noise Ratio (PSNR) values, average mean objective functions values and average CPU rum time values. The experimental analysis shows that the performance of Otsu’s inter-class variance function shows comparatively a better result than Kapur’s and Masi’s entropic method.